A ComfyUI node which copies a given latent's samples tensor along the time axis ((length - 1) // 4) + 1 times to form a longer latent (see EmptyHunyuanLatentVideo's implementation for why this specific number of copies is used) and then prepares a noise_mask tensor of the same shape such that the value of the mask for a given time step is given by the function at https://www.desmos.com/calculator/vhw74mr1vh.

ComfyUI-HunyuanImageLatentToVideoLatent

Copies the given latent's samples tensor along the time axis ((length - 1) // 4) + 1 times to form a longer latent (see EmptyHunyuanLatentVideo's implementation for why this specific number of copies is used). Then, prepares a noise_mask tensor of the same shape such that the value of the mask for a given time step is given by the function shown at https://www.desmos.com/calculator/vhw74mr1vh.

ComfyUI Node: Hunyuan Image Latent To Video Latent

Category

Inputs

Outputs

Extension: ComfyUI-HunyuanImageLatentToVideoLatent

Run ComfyUI workflows in the Cloud!